Method and system for generating edge ai model for edge CCTV

Abstract

A method for generating an edge AI model for edge CCTV is provided. The method comprises the steps of: receiving a first CCTV video analysis result analyzed from at least one edge CCTV among a plurality of edge CCTVs; generating a second CCTV video analysis result using a video analysis AI model on the first CCTV video analysis result; generating a training dataset based on event occurrence time information, video clips, and the second video analysis result corresponding to the first CCTV video analysis result; and training and generating an edge AI model for the edge CCTV based on the training dataset.

Description

Technology Field

[0001] The present invention relates to a method and system for generating an edge AI model for edge CCTV. Background Technology

[0002] Conventional video security services primarily utilized CCTVs equipped with video analysis capabilities. In other words, the method commonly used involved transmitting footage captured by the CCTV to a Video Management System (VMS) control center, where video analysis was then performed.

[0003] However, conventional technologies as described above have a problem in that the cost of building servers for video analysis at VMS control centers continuously increases as the number of CCTVs increases. To address this issue, methods utilizing edge CCTVs equipped with edge modules for video security services are currently being researched.

[0004] In video security services utilizing edge CCTV, a method is used to reduce the burden of video analysis on the VMS control center by performing video analysis on the edge module installed in the CCTV and transmitting the results to the VMS control center. However, current edge CCTVs have a problem in that once an edge AI model for video analysis is installed in the edge module, the installed AI model cannot be changed or modified.

[0005] In addition, recently, to address the aforementioned problems, technologies are being researched to generate multiple edge AI models based on video analysis events or weather conditions, and then apply the edge AI model optimized for those events or conditions to edge CCTVs.

[0006] However, the methods described above have a problem in that they cannot generate and apply an edge AI model optimized for each installed edge CCTV. Therefore, there is a need for technology that automatically generates an edge AI model optimized for each edge CCTV installed at a specific location and automatically updates the edge AI model for the CCTVs. Prior art literature

[0007] Published Patent Application No. 10-2021-0096405 (August 5, 2021) The problem to be solved

[0008] An embodiment of the present invention provides a method and system for generating an edge AI model for an edge CCTV, which automatically generates a training dataset and an edge AI model for an edge CCTV based on the video analysis results of the edge CCTV and the precise video analysis results of the VMS, and loads the generated optimal edge AI model onto the edge CCTV.

[0009] However, the technical problem that this embodiment aims to solve is not limited to the technical problem described above, and other technical problems may exist. means of solving the problem

[0010] As a technical means for achieving the technical problem described above, a method for generating an edge AI model for an edge CCTV according to the first aspect of the present invention comprises: receiving a first CCTV video analysis result analyzed from at least one edge CCTV among a plurality of edge CCTVs; generating a second CCTV video analysis result using a video analysis AI model with respect to the first CCTV video analysis result; generating a training dataset based on event occurrence time information, a video clip, and the second video analysis result corresponding to the first CCTV video analysis result; and training and generating an edge AI model for the edge CCTV based on the training dataset.

[0011] In some embodiments of the present invention, the step of receiving a first CCTV video analysis result analyzed from at least one edge CCTV among the plurality of edge CCTVs may include receiving a first CCTV video analysis result including at least one of an object detection result and a predetermined event detection result from the edge CCTV.

[0012] In some embodiments of the present invention, the step of generating a training dataset based on object and event occurrence time information corresponding to the first CCTV video analysis result, video clips, and the second video analysis result may generate the training dataset by setting the second video analysis result as Ground Truth (GT) information.

[0013] In some embodiments of the present invention, the step of training and generating an edge AI model for the edge CCTV based on the training data set may train and generate the edge AI model when at least one of the cases in which pre-set time information is satisfied and the size of the training data set is greater than or equal to a predetermined size is satisfied.

[0014] Some embodiments of the present invention may further include the steps of: searching for and selecting an edge AI model to be applied to at least one edge CCTV among a plurality of edge CCTVs; and deploying the selected edge AI model to the edge CCTV.

[0015] In some embodiments of the present invention, the step of searching for and selecting an edge AI model to be applied to at least one edge CCTV among the plurality of edge CCTVs may search for and select an edge AI analysis model to be applied to the edge CCTV based on the analysis function and installation size information of the edge AI model.

[0016] In addition, an edge AI model generation system for edge CCTV according to the second aspect of the present invention includes a communication unit that transmits and receives data with a plurality of edge CCTVs, a memory in which a program for learning and generating an edge AI model for the edge CCTV is stored, and a processor that, by executing the program stored in the memory, receives a first CCTV video analysis result analyzed from at least one of the plurality of edge CCTVs through the communication unit, inputs the first CCTV video analysis result into a video analysis AI model to generate a second CCTV video analysis result, constructs a learning data set based on event occurrence time information corresponding to the first CCTV video analysis result, a video clip, and the second video analysis result, and then learns and generates an edge AI model for the edge CCTV based on the learning data set.

[0017] In some embodiments of the present invention, the processor may receive a first CCTV video analysis result including at least one of an object detection result and a predetermined event detection result from the edge CCTV.

[0018] In some embodiments of the present invention, the processor may generate the training data set by setting the second image analysis result as Ground Truth (GT) information.

[0019] In some embodiments of the present invention, the processor can train and generate the edge AI model when the preset time information is satisfied and the size of the training data set is greater than or equal to a predetermined size.

[0020] In some embodiments of the present invention, the processor can search for and select an edge AI model to be applied to at least one edge CCTV among a plurality of edge CCTVs based on the analysis function and installation size information of the edge AI model, and distribute the selected edge AI model to the edge CCTV.

[0021] In addition to this, other methods for implementing the present invention, other systems, and computer-readable recording media for recording a computer program for executing said method may be further provided. Effects of the invention

[0022] According to one embodiment of the present invention described above, an optimized edge AI model can be automatically generated considering the installation environment of each edge CCTV, and the performance of the model can be gradually improved by training the edge AI model based on analysis results obtained from the edge CCTV. Through this, there is an advantage in that rapid tracking and response to crimes, etc., using an edge CCTV-based video security service is possible.

[0023] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art from the description below. Brief explanation of the drawing

[0024] FIG. 1 is a block diagram illustrating the configuration of an edge AI model generation system (100) according to one embodiment of the present invention. FIG. 2 is a flowchart relating to the preparation steps of an edge AI model generation method according to an embodiment of the present invention. FIG. 3 is a flowchart relating to detailed steps in an edge AI model generation method according to an embodiment of the present invention. Specific details for implementing the invention

[0025] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the present invention, and the present invention is defined only by the scope of the claims.

[0026] The terms used in this specification are for describing embodiments and are not intended to limit the invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. The terms "comprises" and / or "comprising" used in this specification do not exclude the presence or addition of one or more other components in addition to the components mentioned. Throughout the specification, the same reference numerals refer to the same components, and "and / or" includes each of the mentioned components and all combinations of one or more. Although terms such as "first," "second," etc., are used to describe various components, these components are not limited by these terms. These terms are used merely to distinguish one component from another. Therefore, the first component mentioned below may be the second component within the technical scope of the invention.

[0027] Unless otherwise defined, all terms used herein (including technical and scientific terms) may be used in a meaning commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0028] Hereinafter, an edge AI model generation system (100) for edge CCTV according to an embodiment of the present invention will be described with reference to FIG. 1.

[0029] FIG. 1 is a block diagram illustrating the configuration of an edge AI model generation system (100) according to one embodiment of the present invention.

[0030] An edge AI model generation system (100) according to one embodiment of the present invention includes an input unit (110), a communication unit (120), a display unit (130), a memory (140), and a processor (150). At this time, the edge AI model generation system in the present invention may be a VMS.

[0031] The input unit (110) generates input data in response to user input of the edge AI model generation system (100). At this time, the user input may include search and selection inputs for edge AI models for CCTVs, and other control inputs such as termination.

[0032] The input unit (110) includes at least one input means. The input unit (110) may include a keyboard, a key pad, a dome switch, a touch panel, a touch key, a mouse, a menu button, etc.

[0033] The communication unit (120) transmits and receives data with multiple CCTVs and performs communication with external devices, such as servers or data collection devices, to transmit and receive data. Such a communication unit (120) may include both a wired communication module and a wireless communication module. The wired communication module may be implemented as a power line communication device, a telephone line communication device, a cable home (MoCA), Ethernet, IEEE1294, an integrated wired home network, and an RS-485 control device. Additionally, the wireless communication module may be composed of a module for implementing functions such as WLAN (wireless LAN), Bluetooth, HDR WPAN, UWB, ZigBee, Impulse Radio, 60GHz WPAN, Binary-CDMA, wireless USB technology and wireless HDMI technology, as well as 5G (5th generation communication), LTE-A (long term evolution-advanced), LTE (long term evolution), and Wi-Fi (wireless fidelity).

[0034] The display unit (130) displays display data according to the operation of the edge AI model generation system (100). For example, the display unit (130) can display information about multiple CCTVs, a list of edge AI models corresponding to each CCTV, information about each edge AI model, and configuration information of a training data set on the screen.

[0035] The display unit (130) includes a liquid crystal display (LCD), a light emitting diode (LED) display, an organic light emitting diode (OLED) display, a micro electro mechanical systems (MEMS) display, and an electronic paper display. The display unit (130) can be combined with the input unit (110) to be implemented as a touch screen.

[0036] The memory (140) stores programs for training and generating edge AI models for edge CCTV. Here, the memory (140) is a general term for non-volatile storage devices and volatile storage devices that retain stored information even when power is not supplied. For example, the memory (140) may include NAND flash memory such as compact flash (CF) cards, SD (secure digital) cards, memory sticks, solid-state drives (SSDs), and micro SD cards, magnetic computer memory devices such as hard disk drives (HDDs), and optical disc drives such as CD-ROMs and DVD-ROMs.

[0037] The processor (150) can execute software, such as a program, to control at least one other component (e.g., a hardware or software component) of the edge AI model generation system (100) and can perform various data processing or operations.

[0038] When the processor (150) receives a first CCTV video analysis result analyzed from at least one edge CCTV among a plurality of edge CCTVs through a communication unit, it inputs the first CCTV video analysis result into a video analysis AI model to generate a second CCTV video analysis result.

[0039] Then, the processor (150) constructs a training data set based on event occurrence time information corresponding to the first CCTV video analysis result, video clips, and the second video analysis result, and then trains and generates an edge AI model for the edge CCTV based on the training data set.

[0040] Meanwhile, in one embodiment of the present invention, the processor (150) may use at least one of machine learning, neural network, or deep learning algorithms as an artificial intelligence algorithm to generate an edge AI model. For example, at least one of machine learning, neural network, or deep learning algorithms may be used as an artificial intelligence algorithm, and examples of neural networks may include models such as CNN (Convolutional Neural Network), DNN (Deep Neural Network), and RNN (Recurrent Neural Network).

[0041] Hereinafter, an edge AI model generation method performed by an edge AI model generation system (100) according to one embodiment of the present invention will be described.

[0042] FIG. 2 is a flowchart relating to the preparation steps of an edge AI model generation method according to an embodiment of the present invention.

[0043] First, the administrator installs an edge CCTV at a specific location (S105).

[0044] In one embodiment of the present invention, an edge CCTV refers to a security terminal equipped with an edge AI module that performs video analysis on a CCTV that performs video recording. Such an edge CCTV records video and generates a first CCTV video analysis result, which is the result of analyzing the recorded video.

[0045] Next, information about the edge CCTV installed in step S105 is registered in the edge AI model generation system (S110).

[0046] In one embodiment, information regarding an edge CCTV may include a unique ID, location, performance information, IP address, etc., for the installed edge CCTV.

[0047] Next, as step S110 ends, the system waits for user input (S115), and if user input is received, determines the input information (S120). At this time, if the user input information is 'Ended' (S120-Ended), the entire process is terminated (S125).

[0048] FIG. 3 is a flowchart relating to detailed steps in an edge AI model generation method according to an embodiment of the present invention.

[0049] If the user's input information in step S120 is determined to be 'Edge AI model application' (S120-Edge AI model application), an Edge AI model to be applied to the Edge CCTV is searched and selected (S130).

[0050] In one embodiment, in step S130, an edge AI model is searched based on the analysis function of the edge AI model, and an edge AI model is selected based on the search results.

[0051] In another embodiment, in step S130, an edge AI model can be searched based on the size of the edge AI model, and an edge AI model can be selected based on the search results.

[0052] In another embodiment, at step S130, all edge AI models stored in the edge AI model generation system can be searched, and an edge AI model can be selected based on the search results.

[0053] Next, the selected edge AI model is distributed to the corresponding edge CCTV (S135), and the edge CCTV that receives the distributed edge AI model is operated after applying it (S140).

[0054] Next, when a first CCTV video analysis result is generated by at least one edge CCTV among the plurality of edge CCTVs (S145), the first CCTV video analysis result is received from the corresponding edge CCTV (S150).

[0055] In this case, the first CCTV video analysis result may include object detection results. Additionally, the first CCTV video analysis result may include event detection results such as fire, flooding, assault, intrusion, loitering, etc. Here, the object detection results and event detection results may be combined and provided as a single piece of information.

[0056] Next, a second CCTV video analysis result is generated using a video analysis AI model based on the first CCTV video analysis result (S155).

[0057] Here, the second CCTV video analysis result refers to a result analyzed more precisely by a high-performance computer with higher specifications than the first edge CCTV video analysis result.

[0058] Next, a training dataset is generated based on event occurrence time information corresponding to the first CCTV video analysis result, video clips, and the second video analysis result (S160).

[0059] In one embodiment, at step S160, a training dataset can be configured by defining the second video analysis result as Ground Truth (GT). Accordingly, a training dataset can be configured such that video clips, objects, and event occurrence time information are set as input data and the second video analysis result is set as output data, thereby allowing an edge AI model for an edge CCTV model to be trained and generated based on the training dataset (S165).

[0060] At this time, in one embodiment of the present invention, step S165 can train and generate an edge AI model when at least one of the cases where the preset time information is satisfied and the size of the training data set is sufficient to satisfy a predetermined size or larger.

[0061] Meanwhile, in one embodiment of the present invention, when a learning data set is configured, the first CCTV video analysis result and the corresponding second CCTV video analysis result can be matched and managed (first case).

[0062] In addition, if multiple first CCTV video analysis results obtained from a single CCTV or multiple different CCTVs are identified as having the same object or event, the multiple first CCTV video analysis results identified as having the same object or event can be grouped. Then, a second CCTV video analysis result can be generated using a video analysis AI model on the grouped first CCTV video analysis results. At this time, the grouped first CCTV video analysis result and the second CCTV video analysis result can be managed by matching (second case).

[0063] In this case, the first case has the advantage of being able to generate and apply an optimized edge AI model by training the edge AI model based on the results analyzed from a single edge CCTV, but the training data set may not be sufficient, and it may take a lot of time to configure the training data set.

[0064] On the other hand, in the case of the second case, where the second CCTV video analysis results generated based on the first CCTV video analysis results grouped as in the second case are configured as a training data set, there is an advantage that the quality of the training data set is maintained while simultaneously enabling faster training of the edge AI model than in the first case.

[0065] Here, multiple first CCTV video analysis results identified as the same object or event are generated by edge CCTVs positioned at different angles or locations within the same site. Therefore, multiple first CCTV video analysis results acquired by multiple edge CCTVs installed in the same location can be simultaneously applied during edge AI model training.

[0066] In addition, in the first case, one second CCTV video analysis result was generated based on one first CCTV video analysis result, but in the second case, since the second CCTV video analysis result is generated in a many-to-one ratio, there is an advantage that faster learning is possible.

[0067] Meanwhile, in the second case, weights may be set and assigned to each second CCTV video analysis result. For example, the first and second CCTV video analysis results in the first case and the first and second CCTV video analysis results in the second case may be compared, and weights may be assigned to each second CCTV video analysis result in the second case according to the degree of agreement.

[0068] Alternatively, a weight may be variably set and assigned to each second CCTV video analysis result based on the number of overlapping intervals and the length of the overlapping intervals of time information regarding the same object or event within a video clip corresponding to the multiple first CCTV video analysis results obtained by multiple edge CCTVs in the second case. That is, even if the number of overlapping intervals in which the same object or event is detected within the video clip of each edge CCTV in the first group and the second group overlaps is the same, if the length of the overlapping interval in the first group is longer, a greater weight may be assigned to the second CCTV video analysis result corresponding to the first group.

[0069] Next, it is determined whether there is user input (S170), and if it is determined that there is no user input (S170-NO), the process returns to step S135. In this case, if a new edge AI model is not created in step S165, steps S135 and S140 are omitted, and the process can be performed starting from step S145.

[0070] In contrast, if a new edge AI model is created, it is executed starting from step S135.

[0071] Additionally, if it is determined that there is user input (S170-YES), the process returns to step S120 and performs an action based on the corresponding input information.

[0072] Meanwhile, in the above description, steps S105 to S170 may be further divided into additional steps or combined into fewer steps according to an embodiment of the present invention. In addition, some steps may be omitted as necessary, and the order between steps may be changed. Furthermore, even if other omitted details are included, the details described in FIG. 1 and FIG. 2 and FIG. 3 apply to each other.

[0073] The embodiments of the present invention described above may be implemented as a program (or application) and stored on a medium to be executed in combination with a server, which is hardware.

[0074] The aforementioned program may include code encoded in a computer language such as C, C++, JAVA, or machine language, which can be read by the computer's processor (CPU) through the computer's device interface, in order for the computer to read the program and execute the methods implemented in the program. Such code may include functional code related to functions that define the necessary functions for executing the methods, and may include control code related to execution procedures necessary for the computer's processor to execute the functions according to a predetermined procedure. Additionally, such code may further include memory reference code regarding where (address) additional information or media necessary for the computer's processor to execute the functions should be referenced in the computer's internal or external memory. In addition, if the processor of the computer needs to communicate with any other computer or server located remotely in order to execute the above functions, the code may further include communication-related code regarding how to communicate with any other computer or server located remotely using the communication module of the computer, and what information or media to transmit or receive during communication.

[0075] The above-mentioned storage medium refers to a medium that stores data semi-permanently and is readable by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, examples of the above-mentioned storage medium include, but are not limited to, ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. That is, the above-mentioned program may be stored on various recording media on various servers that the computer can access, or on various recording media on the user's computer. Additionally, the above-mentioned medium may be distributed across networked computer systems, and computer-readable code may be stored in a distributed manner.

[0076] The steps of the method or algorithm described in connection with embodiments of the present invention may be implemented directly in hardware, implemented as a software module executed by hardware, or implemented by a combination thereof. The software module may reside in RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), Flash Memory, a hard disk, a removable disk, a CD-ROM, or any form of computer-readable recording medium well known in the art to which the present invention belongs.

[0077] Although embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be implemented in other specific forms without altering its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. Explanation of the symbols

[0078] 100: Edge AI Model Generation System 110: Input section 120: Communications Department 130: Display unit 140: Memory 150: Processor

Claims

Claim 1 A method performed by a computer comprises: receiving a first CCTV video analysis result analyzed from at least one edge CCTV among a plurality of edge CCTVs; generating a second CCTV video analysis result using a video analysis AI model with respect to the first CCTV video analysis result; and generating a training dataset based on event occurrence time information corresponding to the first CCTV video analysis result, a video clip, and the second CCTV video analysis result, wherein the training dataset is generated by setting the second CCTV video analysis result as GT (Ground Truth) information. A method for generating an edge AI model for an edge CCTV, comprising the step of training and generating an edge AI model for the edge CCTV based on the above training data set, wherein the first CCTV video analysis result includes a plurality of first CCTV video analysis results of a first group obtained by edge CCTVs of a first group and a plurality of first CCTV video analysis results of a second group obtained by edge CCTVs of a second group, and wherein the step of training and generating the edge AI model includes the step of training and generating the edge AI model by giving greater weight to the second CCTV video analysis result set as the GT (Ground Truth) information corresponding to the first group than to the second CCTV video analysis result corresponding to the second group, even if the number of overlapping sections in which the same object or event is detected within a video clip of each edge CCTV in the first group and the second group is the same, if the length of the overlapping section of the first group is longer. Claim 2 A method for generating an edge AI model for an edge CCTV, wherein, in claim 1, the step of receiving a first CCTV video analysis result analyzed from at least one edge CCTV among the plurality of edge CCTVs is to receive a first CCTV video analysis result including at least one of an object detection result and a predetermined event detection result from the edge CCTV. Claim 3 delete Claim 4 A method for generating an edge AI model for an edge CCTV according to claim 1, wherein the step of training and generating an edge AI model for the edge CCTV based on the training data set is to train and generate the edge AI model when at least one of the cases where preset time information is satisfied and the size of the training data set is greater than or equal to a predetermined size is satisfied. Claim 5 A method for generating an edge AI model for an edge CCTV, further comprising: a step of searching for and selecting an edge AI model to be applied to at least one edge CCTV among a plurality of edge CCTVs in claim 1; and a step of distributing the selected edge AI model to the edge CCTV. Claim 6 A method for generating an edge AI model for an edge CCTV, wherein, in claim 5, the step of searching for and selecting an edge AI model to be applied to at least one edge CCTV among the plurality of edge CCTVs is to search for and select an edge AI analysis model to be applied to the edge CCTV based on the analysis function and installation size information of the edge AI model. Claim 7 A communication unit that transmits and receives data with a plurality of edge CCTVs; a memory storing a program for training and generating an edge AI model for the edge CCTVs; and a processor that, by executing the program stored in the memory, receives a first CCTV video analysis result analyzed from at least one edge CCTV among the plurality of edge CCTVs through the communication unit, inputs the first CCTV video analysis result into a video analysis AI model to generate a second CCTV video analysis result, sets the second CCTV video analysis result as Ground Truth (GT) information, constructs a training data set based on event occurrence time information corresponding to the first CCTV video analysis result, video clips, and the second CCTV video analysis result, and then trains and generates an edge AI model for the edge CCTVs based on the training data set; wherein the first CCTV video analysis result includes a plurality of first CCTV video analysis results of a first group obtained by edge CCTVs of a first group and a plurality of first CCTV video analysis results of a second group obtained by edge CCTVs of a second group, and wherein the processor has a number of overlapping sections in which the same object or event is detected within a video clip of each edge CCTV in the first group and the second group overlaps. An edge AI model generation system for edge CCTV that trains and generates the edge AI model by giving greater weight to the second CCTV video analysis result set with the GT (Ground Truth) information corresponding to the first group than to the second CCTV video analysis result corresponding to the second group, even if identical, when the length of the overlap interval of the first group is longer. Claim 8 An edge AI model generation system for an edge CCTV, wherein, in claim 7, the processor receives a first CCTV video analysis result including at least one of an object detection result and a predetermined event detection result from the edge CCTV. Claim 9 delete Claim 10 An edge AI model generation system for edge CCTV, wherein, in claim 7, the processor learns and generates the edge AI model when pre-set time information is satisfied and when the size of the training data set is greater than or equal to a predetermined size. Claim 11 An edge AI model generation system for edge CCTVs according to claim 7, wherein the processor searches for and selects an edge AI model to be applied to at least one edge CCTV among a plurality of edge CCTVs based on the analysis function and installation size information of the edge AI model, and distributes the selected edge AI model to the edge CCTV.

Images